An AI-based approach for modeling the synergy between radiotherapy and immunotherapy

Abstract

Abstract PULSAR (personalized, ultra-fractionated stereotactic adaptive radiotherapy) is the adaptation of stereotactic ablative radiotherapy towards personalized cancer management, which involves delivering radiation pulses in the ablative range, with intervals separated by weeks or months. The rationale behind this treatment paradigm is that longer intervals between pulses allow for changes in tumors to be utilized in adapting the treatment plan and potentially enhance immune-modulating effects. In our study, we aimed to investigate the interactions between combined PULSAR and PD-L1 blockade immunotherapy based on preclinical studies in syngeneic murine cancer models. Using an LSTM-RNN AI model, we successfully demonstrated that: 1) The LSTM-RNN model can effectively simulate the process of tumor growth and growth delay in a preclinical model, taking into account the combined PULSAR and immunotherapy; 2) The AI model seamlessly integrated various parameters, including pulse interval, radiation dose for each pulse, drug dose, and timing, to predict more effective combinations. Our model excelled in identifying the potential “causal relationship” between tumor growth and the timing of combined treatment, offering two notable advantages: end-to-end learning and prediction. The results of our study showcase significant potential in assisting the implementation of PULSAR and the design of dynamic trials, by harnessing immune-stimulatory effects and ultimately achieving more personalized cancer treatment.